The Advanced Mean-shift Algorithm Based On Texture Match For Fast Real-time Face Detection

Face is a common object in the visual world, which contains a large number of the expression information and the effective identity information of oneself. Face detection technology as the prerequisites of face automatic identification plays a very important role also has significance of the future researches. Face detection comes with Face recognition, but with the advances of technology and the increasing requires, face detection has a wide application in Human Computer Interaction, security surveillance, video conferencing and etc and has attracted a mount of domestic and foreign researchers.The paper firstly introduces and analyzes the classical principle of Mean-shift algorithm. According to the steady feature of the Hue under different light intensity, calculate the probability histogram of the Hue to establish the pattern of the face adapting to the Mean-shift. Then detect the position of the face in the video sequence through the Mean-shift algorithm and analyzes the advantages and disadvantages of the classical Mean-shift algorithm.Secondly, in order to reduce the influence of the similar color of the skin in the Mean-shift algorithm, the paper proposed a method using the match of texture to increase the weight of the trust region. The paper adopted Local Binary Patten (LBP) as the method describing texture information. Form LBP and Hue of of the skin as two dimensional (2D) models in the Mean-shift algorithm. Improve the weight of the trust region which matches the texture during the process of the detecting with Mean-shift algorithm. The results show that the improved method is more successful in real-time face tracking than the original Mean-shift which only adopted the information of the color histogram.Thirdly, the paper proposed the accelerating method of Mean-shift algorithm based on the optical flow algorithm. Pick up the vector of the motion in two frames using the optical flow algorithm to estimate the position of the face. Using this method, the time of searching in one frame can be reduced.At last, it sums up the innovations, the shortage and the work in the future.